Skip to content

[BugFix] VLM RLVR add reward func #851

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
rchardx merged 2 commits into from
Jan 26, 2026
Merged

[BugFix] VLM RLVR add reward func #851

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
44bd31d
VLM RLVR bug fix
PopSoda2002 Jan 26, 2026
229abcc
fix bug
PopSoda2002 Jan 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 9 additions & 2 deletions areal/workflow/vision_rlvr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,8 +8,10 @@
from areal.api.cli_args import GenerationHyperparameters
from areal.api.engine_api import InferenceEngine
from areal.api.io_struct import ModelRequest, ModelResponse
from areal.api.reward_api import AsyncRewardWrapper
from areal.core import workflow_context
from areal.utils import logging, stats_tracker
from areal.utils.dynamic_import import import_from_string
from areal.utils.image import image2base64
from areal.utils.perf_tracer import (
atrace_session_phase,
Expand All @@ -24,9 +26,9 @@
class VisionRLVRWorkflow(RLVRWorkflow):
def __init__(
self,
reward_fn: Callable[..., Any],
reward_fn: Callable[..., Any] | str,
gconfig: GenerationHyperparameters,
tokenizer: PreTrainedTokenizerFast,
tokenizer: PreTrainedTokenizerFast | str,
processor: AutoProcessor | str,
enable_thinking: bool,
):
Expand Down Expand Up @@ -101,6 +103,11 @@ async def _collect_samples(
async def arun_episode(
self, engine: InferenceEngine, data: dict[str, Any]
) -> dict[str, torch.Tensor]:
# NOTE: load reward function dynamically if given as string
if isinstance(self.reward_fn, str):
self.reward_fn = import_from_string(self.reward_fn)
self.async_reward_fn = AsyncRewardWrapper(self.reward_fn)

processor_callable = cast(Callable[..., dict[str, Any]], self.processor)
processed_input = processor_callable(
images=data["images"],
Expand Down